{ "key": "f-nofill-bounds", "approach": "Overlay on N: censored no-fill likelihood. An own order present across a whole step with unchanged remaining quantity (or placed last step and still at full size) was not touched by the arbitrageur, so an unfilled bid at x gives p>=x and an unfilled ask at y gives p<=y (engine.execute_arbitrage sweeps every mispriced level). Interval [lo,hi] = intersect(own no-fill bounds, competitor best bid/ask). N's new PlaceOrders that are asks priced strictly below lo or bids strictly above hi are dropped; everything else passes through unchanged. Acts only when first-step competitor spread s=4.", "tried": [ "Variant A strict (drop askhi), all spreads, tuning 310000..310199: mean_diff -0.347 se 0.288; per s: s1 0.0 (never triggered), s2 -0.611 (0.215), s3 -0.916 (0.221), s4 +0.326 (1.284)", "Variant B inclusive (drop ask<=lo, bid>=hi), all spreads: mean_diff -7.781 se 0.832; every spread negative", "Chosen: variant A, spread set {4}; tuning on s=4 only mean_diff 0.326439 se 1.284461 (n=43); holdout mean_diff -0.099236 se 0.105246: no gain", "Diagnostic: the no-fill/competitor interval was narrower than the competitor spread on 72.9% of steps (holdout mean over runs; tuning 74.3%), yet dropping on it lowers edge: the bound is one step stale (p moves before arb), so N's quotes it removes are mostly profitable retail-catching quotes, not arb losers." ], "next": "Do not promote. The bound information is real (interval narrower than competitor spread ~73% of steps) but as a hard drop rule it is stale by a step; try using the interval midpoint as a soft probability signal fed to N's pricing (shift quotes) rather than dropping, or condition drops on a jump detector.", "spreads_acted": [ 4 ], "off_equals_n": true, "code": "class Strategy(BaseStrategy):\n \"\"\"Overlay f-nofill-bounds: censored no-fill bounds on p.\n\n An own bid at x that sat a whole step untouched means p >= x (else the\n arbitrageur would have sold into it); an untouched ask at y means p <= y.\n Together with the competitor quotes this keeps an interval [lo, hi] (ticks).\n N's new asks priced below lo and new bids priced above hi are dropped.\"\"\"\n\n ACT = (4,)\n STRICT = True\n OFF = False\n\n def __init__(self):\n self._n = _N()\n self._s = None\n self._prev = {}\n self._placed = []\n self.steps = 0\n self.narrow = 0\n self.dropped = 0\n\n def on_step(self, state):\n acts = self._n.on_step(state)\n cb, ca = state.competitor_best_bid_ticks, state.competitor_best_ask_ticks\n if self._s is None:\n if cb is not None and ca is not None:\n self._s = max(1, (ca - cb + 1) // 2)\n else:\n self._s = 0\n lo, hi = None, None\n cur = {}\n placed = self._placed\n for o in state.own_orders:\n cur[o.order_id] = o.remaining_quantity\n q = self._prev.get(o.order_id)\n full = False\n if q is not None:\n full = abs(q - o.remaining_quantity) < 1e-9\n elif o.submitted_step == state.step - 1:\n for (sd, px, qq) in placed:\n if sd == o.side and px == o.price_ticks and abs(qq - o.remaining_quantity) < 1e-9:\n full = True\n break\n if not full:\n continue\n if o.side.value == \"BUY\":\n lo = o.price_ticks if lo is None else max(lo, o.price_ticks)\n else:\n hi = o.price_ticks if hi is None else min(hi, o.price_ticks)\n self._prev = cur\n L = cb if lo is None else (lo if cb is None else max(lo, cb))\n H = ca if hi is None else (hi if ca is None else min(hi, ca))\n self.steps += 1\n if L is not None and H is not None and cb is not None and ca is not None:\n if H - L < ca - cb:\n self.narrow += 1\n out = list(acts)\n if not self.OFF and self._s in self.ACT and (lo is not None or hi is not None):\n keep = []\n for a in out:\n if isinstance(a, PlaceOrder):\n sv = a.side.value\n if sv == \"SELL\" and lo is not None and (a.price_ticks < lo if self.STRICT else a.price_ticks <= lo):\n self.dropped += 1\n continue\n if sv == \"BUY\" and hi is not None and (a.price_ticks > hi if self.STRICT else a.price_ticks >= hi):\n self.dropped += 1\n continue\n keep.append(a)\n out = keep\n self._placed = [(a.side, a.price_ticks, a.quantity) for a in out if isinstance(a, PlaceOrder)]\n return out\n", "code_sha256": "bc5f1d1aeb505ca20ab5f4f6b58037c2a2a2599f9d14fdeb129e3389b439446e", "assembled_sha256": "d05a77cc3c3f1d8cba78a6f596fa354d220a760b5f40540a345f984572427f52", "tuning": { "n": 43, "mean_diff": 0.326439, "se": 1.284461 }, "paired": { "first_seed": 410000, "last_seed": 410499, "n": 500, "complete": true, "incumbent_mean": 104.413753, "mean_diff": -0.099236, "se": 0.105246, "sum_d": -49.61814, "sum_d2": 2768.567977, "retail_diff": 0.056202, "arb_diff": -0.155439, "failures": 0 }, "by_spread": { "1": { "runs": 115, "incumbent_mean": 28.495154, "mean_diff": 0.0, "se": 0.0, "sum_d": 0.0, "sum_d2": 0.0, "retail_diff": 0.0, "arb_diff": 0.0 }, "2": { "runs": 116, "incumbent_mean": 97.766262, "mean_diff": 0.0, "se": 0.0, "sum_d": 0.0, "sum_d2": 0.0, "retail_diff": 0.0, "arb_diff": 0.0 }, "3": { "runs": 138, "incumbent_mean": 128.405451, "mean_diff": 0.0, "se": 0.0, "sum_d": 0.0, "sum_d2": 0.0, "retail_diff": 0.0, "arb_diff": 0.0 }, "4": { "runs": 131, "incumbent_mean": 151.672482, "mean_diff": -0.378764, "se": 0.401829, "sum_d": -49.61814, "sum_d2": 2768.567977, "retail_diff": 0.214512, "arb_diff": -0.593277 } }, "seconds_per_simulation": 0.88 }